I don't use it often. In general, I've found the number of bugs I catch with core.typed doesn't justify the time investment in convincing things to typecheck--my tests generally (not always, of course!) find type issues first. I also tend to do a lot of weird performance-oriented stateful stuff with java interop, which brings me into untyped corners of the library.

That said, I've found core.typed helpful in managing complex state transformations, especially in namespaces which have, say, five or six similar representations of the same logical thing. What do you do when a "node" is a hostname, a logical identifier in Jepsen, an identifier in the database itself, a UID, and a UID+signature pair? Managing those names can be tricky, and having a type system really helps.